Learning Forum

[Amadi]

  One research group here in the College is having issues with LS-DYNA 10.2 .
They are using the MPP version of LS-DYNA 10.2, and when our HPC cluster was upgraded in August from CENTOS7 to Red Hat 8,4 , the 10.2 binary started giving

"*** buffer overflow detected ***

errors . The research group that uses R10.2.0 is not able to switch to a new version,
since the human body model they use only runs correctly using R10.2 with
MPP. Updating the model to a new version is a fairly complex and time
consuming task. This is not something they can do.

I've attached an email from one of our HPC support people showing various things they have tried.

Here is the exact version of LS-DYNA info

Date: 11/08/2022 Time: 20:58:04



___________________________________________________

| |

| Livermore Software Technology Corporation |

| |

| 7374 Las Positas Road |

| Livermore, CA 94551 |

| Tel: (925) 449-2500 Fax: (925) 449-2507 |

| www.lstc.com |

|_________________________________________________|

| |

| LS-DYNA, A Program for Nonlinear Dynamic |

| Analysis of Structures in Three Dimensions |

| Version : mpp s R10.2.0 Date: 03/05/2019 |

| Revision: 135267 Time: 15:49:10 |

| |

| Features enabled in this version: |

| Distributed Memory Parallel |

| FFTW (multi-dimensional FFTW Library) |

| ANSYS License (ans160) |

| |

| Licensed to: |

| |

| Platform : Intel-MPI 2018 Xeon64 |

| OS Level : Linux CentOS 6.5 uum |

| Compiler : Intel Fortran Compiler 16.0 AVX-2 |

| Hostname : caen-vnc-mi16.engin.umich.edu |

| Precision : Single precision (I4R4) |

| SVN Version: 135714 |

| |

| Unauthorized use infringes LSTC copyrights |

|_________________________________________________|





Is there any support we can get with this issue from LSTC .

[Reno Genest]

Hello,

Do you have more information on the error message?

Have you tried with a double precision solver? 

Also, have you tried with a different instruction set like SSE2 instead of AVX2?

You can download different R10.2 solvers here:

https://lsdyna.ansys.com/downloader-filter/

username: user

password: computer

 

Reno.

[Reno Genest]

Hello,

I checked with a colleague and here is his reply and solution for the buffer overflow detected error:

“A simple solution in advance:
Just run this command
$ sudo sysctl -w kernel.pid_max=999999
(innocuous command, no need to reboot)
and all versions will run as before on older CentOS – without any downside.

The reason is a bug in old Intel compiler which shows up if process IDs gets larger than 7 digits (=1million) – but only if the job is run in background

Here is a longer info:
regarding “buffer overflow detected”:
unfortunately this is a known issue with binaries built with Intel-Fortran-16.0 up to 16.3
and process IDs > 999.999 :
https://community.intel.com/t5/Intel-Fortran-Compiler/redirecting-stdout-from-ifort-16-0-1-compiled-program-results-in/td-p/1106715
https://community.intel.com/t5/Intel-Fortran-Compiler/redirecting-stdout-from-ifort-16-0-1-compiled-program-results-in/m-p/1106727#M128232
The error occurs with any version under these conditions:
- build with ifort16.0.3
- stdout is redirected to file (eg in batch jobs)
- current process ID has 7 digits  or more

The limit of the max process-ID can be found for any OS here:
$ cat /proc/sys/kernel/pid_max

the typical value before RHEL/CentOS 8.x was:
$ cat /proc/sys/kernel/pid_max
32768
(just 5 digits)

But since RHEL/CentOS8.x it has 7 digits:
$ cat /proc/sys/kernel/pid_max 
4194304

You may choose one of the following workarounds
a) use a binary which is not built with Ifort16
b) use a binary which is more recent than August 2021
c) switch the current PID to a lower number (requires sudo permissions), eg:
   $ sudo echo 1 > /proc/sys/kernel/ns_last_pid
d) limit the max PID to 6 digits (no need to reboot):
   $ sudo sysctl -w kernel.pid_max=999999

Solution d) is favorized by most sites.

”

 

Reno.